The center of gravity in chips is tilting hard toward Asia today — with sovereign-scale money, a Chinese surprise at the top of the supercomputing chart, and a memory squeeze that won't quit.
South Korea Goes All In
South Korea opened the week swinging. President Lee Jae Myung unveiled three "mega-projects" pledging up to $1 trillion to anchor the country's next growth phase around semiconductors and AI, aiming to lock in what he called an overwhelming lead. The ambition wasn't free, though. Shares in Samsung Electronics and SK Hynix — two of the world's biggest chipmakers — slid sharply after reports the pair are preparing roughly $1.3 trillion in new investment plans. Investors, it seems, want to see the returns before they cheer the spending.
China Closes the Gap
China supplied the day's most striking headline: a system called LineShine has been ranked the fastest supercomputer in the world — and it did it without a single GPU, a pointed statement about alternative paths to high performance amid chip restrictions.
The momentum extends to the market. Nvidia, long dominant in AI chips, is visibly losing ground in China, where Huawei is surging into the gap. Meanwhile Baidu's AI chip arm, Kunlunxin, is steering toward a Hong Kong IPO at a target valuation of $50 billion. Its pitch carries an unusual twist: prospective IPO investors are reportedly being asked to also buy its chips — blurring the line between shareholder and customer in a way that underscores how badly Chinese players want domestic demand locked in.
Nvidia Plays Offense Elsewhere
Even as China slips, Nvidia is planting flags across the wider Asia-Pacific. It's backing Firmus, a Tasmania-based startup, in a deal anchoring a $20 billion-plus AI data center in Indonesia — one of the largest digital-infrastructure pushes the region has seen. The message: if one market cools, build the next one.
Custom Silicon Heats Up
The in-house chip race took a flavorful turn. OpenAI and Broadcom unveiled a custom AI chip nicknamed "Jalapeño," built to run the next wave of large language models. The headline claim is cost: roughly 50% savings versus the status quo. For a company whose compute bills scale with every model it ships, halving the per-chip economics is the kind of number that reshapes strategy — and pressures the merchant-silicon incumbents.
The Memory Crunch Gets Real
If there's a through-line tying the spending frenzy together, it's memory. Makers of memory chips — the components that store data feeding AI systems — are enjoying a windfall as the AI boom sends prices soaring. The flip side landed at ISC 2026, where Lenovo warned of a permanent "RAMageddon": cheap, abundant memory may not be coming back, and shortage is becoming the "new normal." Buyers of everything from laptops to data centers should plan accordingly.
Steady Hand
Amid the drama, one bellwether stayed calm. TSMC, the world's largest contract chipmaker, posted a rise in monthly sales against a backdrop of firm, steady chip demand — a quiet reminder that beneath the trillion-dollar pledges and supercomputing upsets, the orders are still flowing.
The takeaway: governments and giants are committing historic sums, China is proving it can compete on its own terms, and the constraint that may matter most isn't ambition — it's memory.